New wells production rate forecast quality improvement due to calculated values achievement for objects with different geological, geophysical and technological parameters

UDK: 681.518:622.276.346
DOI: 10.24887/0028-2448-2023-8-128-130
Key words: new well production rate, data analysis, machine learning, reservoir engineering, risk management
Authors: K.V. Kudashov (Rosneft Oil Company, RF, Moscow), V.E. Antsupov (Rosneft Oil Company, RF, Moscow), D.A. Akimova (Rosneft Oil Company, RF, Moscow)

This paper is focused on building a model using actual data that links the degree of achievement of the planned rates of new wells with geological, geophysical and technological factors.

The goal of this work is to improve the quality of new well production rate forecast by implementing an additional adjustment, derived from the solution of the regression problem to predict the difference between actual and planned (calculated) new well production rates based on given properties of drilling targets and completion technology.

Two models based on different machine learning tools were built to solve the stated problem. The difference between planned and actual production rates was chosen as a target function. The first model is a classification model. In this case, the classification problem was solved and the sign of the target function was predicted, then the regression problem was solved for each of the two ranges. The second model is the clustering model. Its main idea was to divide the objects under study into clusters. Then, in each of the clusters, the regression problem was solved. The models were trained on the data on the new wells of Rosneft Oil Company PJSC for the period from 2017 to 2021.

As the result of the study, the above two models were used to predict the difference between calculated and actual rates for each of the planned new wells in the period from 2023 to 2027. Furthermore, prediction quality was tested on results obtained from new wells that were drilled within four months of 2023. This test showed an increase in forecast quality for 658 wells sample average rate from 95 to 97−99 %.

 

References

1. Hastie T., Tibshirani R., Friedman J., The Elements of statistical learning. Data mining, inference and prediction, New York: Springer-Verlag, 2009, 745 p.

2. Kugaevskikh A.V., Muromtsev D.I., Kirsanova O.V., Klassicheskie metody mashinnogo obucheniya (Classical machine learning methods), St. Petersburg: Publ. of ITMO University, 2022, 53 p.

3. Limanovskaya O.V., Alfer'eva T.I., Osnovy mashinnogo obucheniya (Fundamentals of machine learning), Ekaterinburg: Publ. of Ural University, 2020, 88 p.

4. Azbukhanov A.F., Kostrigin I.V., Bondarenko K.A. et al., Selection of wells for hydraulic fracturing based on mathematical modeling using machine learning methods (In Russ.), Neftyanoe khozyaystvo = Oil Industry, 2019, no. 11, pp. 38-42, DOI: https://doi.org/10.24887/0028-2448-2019-11-38-42

5. Akimova D.A., Issledovanie zavisimosti dostizheniya planiruemykh debitov novykh skvazhin ot geologo-geofizicheskikh i tekhnologicheskikh faktorov (Study of the dependence of achieving the planned flow rates of new wells on geological, geophysical and technological factors): graduation qualification work, Proceedings of the 65th All-Russian Scientific Conference of the Moscow Institute of Physics and Technology in honor of the 115th anniversary of L.D. Landau, Moscow, 3-8 April 2023, Moscow: Publ. of MIPT, 2023.



Attention!
To buy the complete text of article (Russian version a format - PDF) or to read the material which is in open access only the authorized visitors of the website can. .